Multi SVR Approach for Fault Location in Multi-terminal HVDC Systems

Abstract

In this paper, a method based on machine learning strategies is proposed to address fault location problem of multiterminal HVDC systems. Support vector regression (SVR) is employed to locate different faults in the system. The SVR is trained using extracted signatures of different voltage and current signals by utilizing wavelet transform. Different combinations of voltage and current signals of the both line’s sides are taken into consideration in order to find the best option from accuracy of the method aspect. The best setting for the SVR parameters are derived. Also, accuracy of the method is investigated for various lengths of the analyzed window. Two approaches are considered for applying the method to the system. The first one is the regular approach where an SVR is used for whole the line’s length. A novel approach named multi-SVR approach is proposed here where, the transmission line is sectionalized and separate SVRs are applied to every section. It is shown that performance of the method is enhanced using the multi-SVR approach rather than the single SVR as every SVR focuses on smaller areas. The method performance is assessed using different simulations of a light HVDC system in different conditions.